Performance Optimisation - ECMWF · •On Performance Optimisation and Productivity •Promoting best practices in parallel programming •Providing Services •Precise understanding

EU H2020 Centre of Excellence (CoE) 1 October 2015 – 31 March 2018

Grant Agreement No 676553

Performance Optimisation and Productivity

Nick Dingle [email protected]

Why?

• Complexity of machines and codes

Frequent lack of quantified understanding of actual behaviour

Not clear most productive direction of code refactoring

• Important to maximize efficiency (performance, power) of compute intensive applications and productivity of the development efforts

What?

• Parallel programs, mainly MPI/OpenMP• Although also CUDA, OpenCL, OpenACC, Python, …

2

Motivation

28/10/2016 ECMWF 17th Workshop on HPC in Meteorology [email protected]

POP CoE

• A Centre of Excellence• On Performance Optimisation and Productivity

• Promoting best practices in parallel programming

• Providing Services • Precise understanding of application and system behaviour

• Suggestion/support on how to refactor code in the most productive way

• Horizontal• Transversal across application areas, platforms, scales

• For academic and industrial codes and users

• FREE !


? Parallel Application Performance Audit Report• Primary service

• Identify performance issues of customer code (at customer site)

• Small effort (< 1 month)

! Parallel Application Performance Plan Report• Follow-up on the audit service

• Identifies the root causes of the issues found andqualifies and quantifies approaches to address them

• Longer effort (1-3 months)

Proof-of-Concept Software Demonstrator• Experiments and mock-up tests for customer codes

• Kernel extraction, parallelisation, mini-apps experiments to showeffect of proposed optimisations

• 6 months effort

Services provided by the CoE


• Application Structure • (if appropriate) Region of Interest• Scalability Information• Application Efficiency

• E.g. time spent outside MPI

• Load Balance• Whether due to internal or external factors

• Computational Performance• Identification of areas for improvement

• Communications• E.g. sensitivity to network performance

• Summary and Recommendation

5

Outline of a typical audit report


The process …

When?

October 2015 – March 2018

How?

• Apply

• Fill in small questionnairedescribing application and needshttps://pop-coe.eu/request-service-form

• Questions? Ask [email protected]

• Selection/assignment process

• Install tools @ your production machine

• Interactively: Gather data Analysis Report


https://pop-coe.eu/request-service-form

mailto:[email protected]

Customer feedback

7

• What was the quality of their answers?• How responsive have the POP experts been to your questions or concerns about the analysis and the report?28/10/2016 ECMWF 17th Workshop on HPC in Meteorology [email protected]

• Who?• BSC (coordinator), ES

• HLRS, DE

• JSC, DE

• NAG, UK

• RWTH Aachen, IT Center, DE

• TERATEC, FR

A team with

• Excellence in performance tools and tuning

• Excellence in parallel programming models and practices

• Research and development background AND proven commitment in application to real academic and industrial use cases

8

Partners


Target customers

• Code developers

• Assessment of detailed actual behaviour

• Suggestion of most productive directions to refactor code

• Users

• Assessment of achieved performance in specific production conditions

• Possible improvements modifying environment setup

• Evidence to interact with code provider

• Infrastructure operators

• Assessment of achieved performancein production conditions

• Possible improvements from modifying environment setup

• Information for time computer time allocation processes

• Training of support staff

• Vendors

• Benchmarking

• Customer support

• System dimensioning/design


Tools

• Install and use already available monitoring and analysis technology

• Open-source toolsets

• Extrae + Paraver

• Score-P + Cube + Scalasca/TAU

• Dimemas, Extra-P

• SimGrid

• Commercial toolsets (if available at customer site)

• Intel tools

• Cray tools

• Allinea tools


Code Audit Examples


12

GITM (Cefas)

• An offline particle tracking model code

• Written in Fortran with OpenMP

• Key audit results:• Current performance: 16 threads offer 5x speed-up vs 1 thread

• Load imbalance amongst OpenMP threads

• Maximise opportunities for vectorisation by aligning arrays and refactoring Fortan array operations to use DO loops

• Large number of floating-point divisions

28/10/2016

• Extracted from ECMWF's Integrated Forecasting System (IFS) code

• Mixed Fortran/C++ code with hybrid MPI+OpenMP

• Key audit results:

13

dwarf-D-ellipticSolver-GCR (ECMWF)

• Good scalability as communication pattern introduces little overhead from synchronisation or serialisation

• Minor load imbalance due to variability in IPC across nodes

• Opportunities to further increase performance as IPC is 0.8 on average

28/10/2016

• Undertaken as part of the ESCAPE (Energy-efficient Scalable Algorithms for Weather Prediction at Exascale) project

• Written in Fortran with MPI

• Traces gathered by user and analysed by POP

• Key audit findings:

14

NEMO (Atos)

• Good computational load balance • Observed superlinear speed-up of

analysed region of the program, which was due to improved cache efficiency when strong scaling

28/10/2016

Other POP activities

• Customer advocacy• Gather customers feedback, ensure satisfaction, steer activities

• Sustainability• Explore business models

• Training• Best practices on the use of the tools and programming models


28/10/2016 16

Contact:https://[email protected]

This project has received funding from the European Union‘s Horizon 2020 research and innovation programme under grant agreement No 676553.

Performance Optimisation and Productivity A Centre of Excellence in Computing Applications

Performance Optimisation - ECMWF · •On Performance Optimisation and Productivity •Promoting best practices in parallel programming •Providing Services •Precise understanding

Documents